How do you check that an AI answer matches your current policy?
You write the rules, such as matching current prices and return windows, naming a source, and giving no legal, medical, or tax advice. Your agent submits each answer to Vindex with the policy text it drew from, and an independent model rules on it, naming any rule it did not pass.
You write the rules
Office Manager or Head of Support, and Compliance
Your agent does the work
Answers to customer and staff questions drawn from your policies, price lists, and help articles.
Vindex rules on it
An independent model checks each output against the rule version in force and names any rule it did not pass.
Each branch checks the others. The agent never reviews its own output.
The rules you write
These are examples to adapt. Your company drafts, versions, and owns every rule, in plain language or from public rules. An agent can draft a rule. Only a person makes it law.
Matches current policy
Every price, warranty term, and return window in the answer matches the policy text supplied with it.
Source named
The answer names the document and section it came from.
Coverage claims
The answer does not say a service is covered unless the supplied plan or benefit record says so.
Private details
The answer contains no personal details of anyone other than the person asking.
Scope
The answer gives no legal, medical, or tax advice.
Submit the policy text with the answer so the rule has something to compare against. Vindex checks the answer against the rules you wrote. It does not decide what a privacy law requires.
